Barking is perhaps the most characteristic formudof vocalization in dogs; however, very little is known aboutudits role in the intraspecific communication of this species.udBesides the obvious need for ethological research, both inudthe field and in the laboratory, the possible informationudcontent of barks can also be explored by computerizedudacoustic analyses. This study compares four differentudsupervised learning methods (naive Bayes, classificationudtrees, k-nearest neighbors and logistic regression) combinedudwith three strategies for selecting variables (alludvariables, filter and wrapper feature subset selections) toudclassify Mudi dogs by sex, age, context and individualudfrom their barks. The classification accuracy of the modelsudobtained was estimated by means of K-fold cross-validation.udPercentages of correct classifications were 85.13 %udfor determining sex, 80.25 % for predicting age (recodifiedudas young, adult and old), 55.50 % for classifying contextsud(seven situations) and 67.63 % for recognizing individualsud(8 dogs), so the results are encouraging. The best-performingudmethod was k-nearest neighbors following audwrapper feature selection approach. The results for classifyingudcontexts and recognizing individual dogs were betterudwith this method than they were for other approachesudreported in the specialized literature. This is the first timeudthat the sex and age of domestic dogs have been predictedudwith the help of sound analysis. This study shows that dogudbarks carry ample information regarding the caller’sudindexical features. Our computerized analysis providesudindirect proof that barks may serve as an important sourceudof information for dogs as well.
展开▼
